A Dynamic Time-Series Model of Oxygen Consumption during Paediatric Cardiopulmonary Bypass

Abstract

AbstractBackgroundDuring cardiopulmonary bypass (CPB), maintaining adequate oxygen consumption (VO2i) can only be achieved indirectly either by modifying oxygen delivery (DO2i) through its component parts or by modulating metabolic demand through altering body temperature. The body reacts to these actions by changing OER and consequently VO2i. Understanding the body’s adaptive OER dynamics can elucidate its oxygen consumption goals during CPB and help improve our ability to safely manage the patient’s journey.MethodsAn autoregressive, integrated time-series model was trained on granular perfusion data from 879 paediatric patients (age: newborn to 18 years old) undergoing 963 CPB operations, with the outcome variable being the minute-by-minute changes in the logit transformation of OER. Variables were cardiac index, haemoglobin concentration, oxygen saturation of arterial haemoglobin and temperature. An explicit ‘disequilibrium term group’ was also included, proportional to the difference between the logarithm of VO2i and logarithm of a ‘latent’ (i.e. unobserved) oxygen demand - or ‘target’ VO2(tVO2i) - term, with the logarithm of tVO2i assumed to be a linear function of body temperature (the Van’t Hoff model). The trained time-series models were studied using permutation-based variable importance, deterministic and stochastic simulations, and subgroup analysis by acute kidney injury (AKI) grade and by temperature.ResultsModel coefficients are consistent with an adaptive OER response to keep VO2i in line with tVO2i, according to body temperature. This adaptation consists of a primary rapid response for 5-10 minutes, and a secondary slow response that is estimated to last up to several hours. The model reproduces the hyperbolic shape of DO2i-VO2i curves - first published in 1982 - as an artefact of insufficient wait times between equilibrium-state transitions. Asymptotically, however, the model converges to a piecewise linear relationship between DO2i and VO2i, with supply-independence of oxygen consumption occurring above a threshold DO2i. Subgroup analysis by temperature suggests that the dependence of tVO2i on temperature (expressed as Q10) may be significantly stronger at low temperatures (< 28C) than at high temperatures (> 28C).ConclusionsThis study proposes a physiologically plausible model of OER changes during CPB that is consistent with past experimental data. While during CPB, under-oxygenation is the dominant risk in the long term, slow adaptation of OER during CPB creates short-term opportunities for over-oxygenation following significant changes in variables such as cardiac index. The model provides well-defined values for tVO2i at a given temperature, paving the way for further research into the effects of over- and under-oxygenation during CPB on postoperative outcomes such as AKI, and hence improvements in goal-directed perfusion protocols.Clinical PerspectiveWhat Is New?This study is the first to present a data-driven, analytical framework for predicting OER changes in response to clinical interventions during CPB.Changes in the components of oxygen delivery cause an adaptive OER response to keep oxygen consumption in line with oxygen demand, according to body temperature.The dependence of oxygen demand on temperature decreases as temperature increases towards normothermia, inconsistent with the accepted Van’t Hoff equation.Children developing AKI exhibit a dampened response to changes in haemoglobin during CPB, with this dampening of response intensifying with AKI severity.What Are the Clinical Implications?This proposed, dynamic model of OER provides a novel framework for goal-directed perfusion by identifying periods of over- and under-oxygenation.The observed, dampened response to haemoglobin changes in patients that develop AKI can be the foundation of an intraoperative tool for early diagnosis of at-risk patients.